Fork me on GitHub
#data-science
<
2018-12-27
>
Daniel Slutsky16:12:32

Here are some very basic course notes of doing machine learning in clojure with Smile. http://viewer.gorilla-repl.org/view.html?source=github&amp;user=clojure-data-science-course&amp;repo=examples&amp;path=src/examples/basic_machine_learning.clj Maybe you may find it useful. I would love to hear any thoughts or comments.

alan22:12:57

Hey, seems interesting, besides I wanted to start wrapping smile in Clojure and some help would be great!

alan22:12:45

Anyway, I'll try to take a better look at the notes tomorrow and come back with a few comments 😀

Daniel Slutsky23:12:28

Hi @UAE4G28HX, wonderful - I would love to discuss it. We, too, are beginning to write some Smile wrapper.

alan09:12:06

I have a few comments: - Gorilla REPL is nice, but is pretty much abandoned, you might want to use clojupyter https://github.com/clojupyter/clojupyter instead - Instead of Specter there's huri https://github.com/sbelak/huri that has much more functionality

alan09:12:48

About wrappers I had an exchange (https://github.com/haifengl/smile/issues/219) with the creator of SMILE and he is willing to add a Clojure API to SMILE, provided that someone else writes wrappers. I already did something similar with XGBoost (https://gitlab.com/alanmarazzi/clj-boost) which might be interesting for the course as well 😄

Daniel Slutsky16:12:01

Thanks! Yes, clojupyter is quite useful. On a local clojure-data-science course that we are having these days, we began doing some tutorials in clojupyter, but then switched to gorilla-repl, which seems more beginner friendly (e.g., installation is much simpler, and there is less need to cope with kernel restarts and explain such delicate situations). Clj-boost is great and useful - thank you for your work on that! In the long run, it seems important to create some unified clojure API around several jvm-based machine learning libraries, like smile, weka, liblinear-java, xgboost, dl4j. This will allow us to conduct unified experiments with systematic comparisons of different methods. Here one may learn from the experience of R (carret) and Python (scikit-learn). If you wish, I would love to have some online meeting (skype?), and see if we can cooperate. I have also been discussing this with @U086BARD5, who has given some thought to similar questions and may be able to help.

Daniel Slutsky16:12:52

May I suggest that we open a dedicate slack channel for discussing machine learning wrappers?

alan16:12:00

I second both the dedicated channel and the Skype call! I'm not sure I'll be available before 2019, but I'll try. I'm in Italy, where are you located (just a matter of time zones)?

alan16:12:16

And I'm very happy someone finds clj-boost useful!

Daniel Slutsky19:12:02

Thanks , continuing on a new channel 🙂.